simplify

Signed-off-by: Gyuho Lee <[email protected]>

components/accelerator/nvidia/bad-envs/component.go

@@ -8,12 +8,13 @@ import (
 
 	"github.com/leptonai/gpud/components"
 	bad_envs_id "github.com/leptonai/gpud/components/accelerator/nvidia/bad-envs/id"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_query "github.com/leptonai/gpud/components/accelerator/nvidia/query"
 	"github.com/leptonai/gpud/components/query"
 	"github.com/leptonai/gpud/log"
 )
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 
 	cctx, ccancel := context.WithCancel(ctx)

components/accelerator/nvidia/clock-speed/component.go

@@ -9,6 +9,7 @@ import (
 
 	"github.com/leptonai/gpud/components"
 	nvidia_clock_speed_id "github.com/leptonai/gpud/components/accelerator/nvidia/clock-speed/id"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_query "github.com/leptonai/gpud/components/accelerator/nvidia/query"
 	nvidia_query_metrics_clockspeed "github.com/leptonai/gpud/components/accelerator/nvidia/query/metrics/clock-speed"
 	"github.com/leptonai/gpud/components/query"
@@ -17,7 +18,7 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 
 	cctx, ccancel := context.WithCancel(ctx)

components/accelerator/nvidia/bad-envs/config.go → components/accelerator/nvidia/common/config.go

@@ -1,4 +1,4 @@
-package badenvs
+package common
 
 import (
 	"database/sql"
@@ -10,6 +10,10 @@ import (
 type Config struct {
 	Query query_config.Config `json:"query"`
 
+	ToolOverwrites
+}
+
+type ToolOverwrites struct {
 	NvidiaSMICommand         string `json:"nvidia_smi_command"`
 	NvidiaSMIQueryCommand    string `json:"nvidia_smi_query_command"`
 	IbstatCommand            string `json:"ibstat_command"`

components/accelerator/nvidia/ecc/component.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_ecc_id "github.com/leptonai/gpud/components/accelerator/nvidia/ecc/id"
 	nvidia_query "github.com/leptonai/gpud/components/accelerator/nvidia/query"
 	nvidia_query_metrics_ecc "github.com/leptonai/gpud/components/accelerator/nvidia/query/metrics/ecc"
@@ -17,7 +18,7 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 
 	cctx, ccancel := context.WithCancel(ctx)

components/accelerator/nvidia/error-xid-sxid/component.go

@@ -9,6 +9,7 @@ import (
 
 	"github.com/dustin/go-humanize"
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_error_xid_sxid_id "github.com/leptonai/gpud/components/accelerator/nvidia/error-xid-sxid/id"
 	nvidia_query "github.com/leptonai/gpud/components/accelerator/nvidia/query"
 	nvidia_xid_sxid_state "github.com/leptonai/gpud/components/accelerator/nvidia/query/xid-sxid-state"
@@ -18,7 +19,7 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 
 	// this starts the Xid poller via "nvml.StartDefaultInstance"
@@ -44,7 +45,7 @@ func New(ctx context.Context, cfg Config) components.Component {
 var _ components.Component = (*component)(nil)
 
 type component struct {
-	cfg     Config
+	cfg     nvidia_common.Config
 	rootCtx context.Context
 	cancel  context.CancelFunc
 	poller  query.Poller

components/accelerator/nvidia/error/component.go

@@ -7,14 +7,15 @@ import (
 	"time"
 
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_query "github.com/leptonai/gpud/components/accelerator/nvidia/query"
 	"github.com/leptonai/gpud/components/query"
 	"github.com/leptonai/gpud/log"
 )
 
 const Name = "accelerator-nvidia-error"
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 
 	cctx, ccancel := context.WithCancel(ctx)

components/accelerator/nvidia/error/xid/component.go

@@ -9,6 +9,7 @@ import (
 	"time"
 
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_component_error_xid_id "github.com/leptonai/gpud/components/accelerator/nvidia/error/xid/id"
 	nvidia_query_nvml "github.com/leptonai/gpud/components/accelerator/nvidia/query/nvml"
 	nvidia_query_xid "github.com/leptonai/gpud/components/accelerator/nvidia/query/xid"
@@ -17,7 +18,7 @@ import (
 	"github.com/leptonai/gpud/log"
 )
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 	setDefaultPoller(cfg)
 

components/accelerator/nvidia/error/xid/component_output.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_component_error_xid_id "github.com/leptonai/gpud/components/accelerator/nvidia/error/xid/id"
 	nvidia_query_nvml "github.com/leptonai/gpud/components/accelerator/nvidia/query/nvml"
 	nvidia_query_xid "github.com/leptonai/gpud/components/accelerator/nvidia/query/xid"
@@ -223,7 +224,7 @@ var (
 )
 
 // only set once since it relies on the kube client and specific port
-func setDefaultPoller(cfg Config) {
+func setDefaultPoller(cfg nvidia_common.Config) {
 	defaultPollerOnce.Do(func() {
 		defaultPoller = query.New(
 			nvidia_component_error_xid_id.Name,

components/accelerator/nvidia/fabric-manager/config.go

@@ -4,6 +4,7 @@ import (
 	"database/sql"
 	"encoding/json"
 
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	fabric_manager_log "github.com/leptonai/gpud/components/accelerator/nvidia/query/fabric-manager-log"
 	query_config "github.com/leptonai/gpud/components/query/config"
 	query_log_common "github.com/leptonai/gpud/components/query/log/common"
@@ -15,11 +16,7 @@ import (
 type Config struct {
 	Query query_config.Config     `json:"query"`
 	Log   query_log_config.Config `json:"log"`
-
-	NvidiaSMICommand         string `json:"nvidia_smi_command"`
-	NvidiaSMIQueryCommand    string `json:"nvidia_smi_query_command"`
-	IbstatCommand            string `json:"ibstat_command"`
-	InfinibandClassDirectory string `json:"infiniband_class_directory"`
+	nvidia_common.ToolOverwrites
 }
 
 func ParseConfig(b any, dbRW *sql.DB, dbRO *sql.DB) (*Config, error) {

components/accelerator/nvidia/gpm/component.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_query_metrics_gpm "github.com/leptonai/gpud/components/accelerator/nvidia/query/metrics/gpm"
 	nvidia_query_nvml "github.com/leptonai/gpud/components/accelerator/nvidia/query/nvml"
 	components_metrics_state "github.com/leptonai/gpud/components/metrics/state"
@@ -19,7 +20,7 @@ import (
 
 const Name = "accelerator-nvidia-gpm"
 
-func New(ctx context.Context, cfg Config) components.Component {
+func New(ctx context.Context, cfg nvidia_common.Config) components.Component {
 	cfg.Query.SetDefaultsIfNotSet()
 	setDefaultPoller(cfg)
 

components/accelerator/nvidia/gpm/component_output.go

@@ -8,6 +8,7 @@ import (
 	"sync"
 
 	"github.com/leptonai/gpud/components"
+	nvidia_common "github.com/leptonai/gpud/components/accelerator/nvidia/common"
 	nvidia_query_nvml "github.com/leptonai/gpud/components/accelerator/nvidia/query/nvml"
 	components_metrics "github.com/leptonai/gpud/components/metrics"
 	"github.com/leptonai/gpud/components/query"
@@ -96,7 +97,7 @@ var (
 )
 
 // only set once since it relies on the kube client and specific port
-func setDefaultPoller(cfg Config) {
+func setDefaultPoller(cfg nvidia_common.Config) {
 	defaultPollerOnce.Do(func() {
 		defaultPoller = query.New(
 			Name,